Underwater image super-resolution reconstruction method and dynamic perception modulation network thereof

The invention provides a dynamic perception modulation network which comprises a shallow layer extraction module, a dynamic weight acquisition module, an up-sampling reconstruction module and a plurality of deep layer extraction modules arranged in series, and each deep layer extraction module comprises a Patch Embedding block, a Mixed Transformer block and a mixed attention fusion module. According to the dynamic perception modulation network provided by the invention, the robustness of detail information reconstruction is improved through embedded structure prior, and meanwhile, an excellent image super-resolution reconstruction effect is obtained on the basis of keeping relatively low calculation requirements; when the method is used for underwater image reconstruction, the method shows good PSNR and SSIM performance. 本发明提供了一种动态感知调制网络，包括浅层提取模块、动态权重获取模块、上采样重建模块以及以串行排列的多个深层提取模块，其中每个深层提取模块分别包括Patch Embedding块、Mixed Transformer块和混合注意力融合模块。本发明提出的动态感知调制网络，通过嵌入结构先验，提升细节信息重建的鲁棒性，同时在保持较低的计算需求基础上获得优异的图像超分辨率重构效果；在用于水